×××× ××××× 21 ××××× 2003, 19:42, ×××× ×× ××× Gilad Ben-Yossef: > On Sunday 21 December 2003 18:37, Oded Arbel wrote: > > HURD offers something very interesting in this areana: you won't need UML > > with HURD because each user can run her own drivers/filesystems/etc or > > even a full kernel on a running system w/o affecting other users. > > Which is exactly what UML does. The fact the in HURD it will be your_kernel > talking to the microkernel and on Linux it's UML talking to the host kernel > is moot
It is if you want to run a full OS, but the HURD allows you to run just a filesystem (and pretend its root - a-la chroot), or just a specific driver or any combinartion of hirds that you feel like using, and most times you don't even need super-user premission to do that. > > > Is the situation considerably better in x86-based BSD systems? > > > > Not AFAIK. marginly better I might say. > > Technically, that's wrong. Darwin for example is a BSD system running on > top of the Mach microkernel, the same one that HURD used in the begining of > the project. And the situation with Darwin is marginly better then in Linux regarding most of the problems Ive mentioned. HURD has a bit different approach then Darwin to how the kernel works with the rest of the system, and I think the situation there is much better (again, not taking into account amount, quality and availability of hardware drivers as this is clearly not a fair comparison). > > There are tons of problems, one that comes to mind now (because I just > > ran into it today) is that if a process is blocked on IO, inside the > > kernel, nothing you can do in user space can free it. you can't interrupt > > it or even kill -9 it. if you can't fix the problem at the root, you > > might as well reboot. > > But have you ever asked yourself why this is so? <snip> > But now consider what happens if you are doing IO. I mean real IO here - > talking to some hardware or such. As the kernel, you started handling the > request and sent some instructions to the hardware, like wrote some stuff > to a region of memory the card reads via DMA for example. Yes, but mostly the problems are with simple stuff like PIPEs, network sockets and such, and I think that the kernel should make allowences for those, or atleast for the KILL signal. > And after this too long a speech, maybe all you need is to add "soft" to > the NFS volume mount options? :-)))) because it doesn't help any- it blocks as well, especially if portmap isn't running.. besides, why soft isn't the default ? we know that the abstraction doesn't work so why try to enforce it ? -- Oded ::.. No program done by an undergrad will work after she graduates. ================================================================To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]
